The grid id and URL omitted the model, so two models sharing a cycle produced
the same `/api/wind/grid/<id>.bin`. Because the grid route defaults to `gfs`,
an IFS grid request resolved against the GFS cache: when both cycles matched it
silently served the GFS field as IFS, and when they differed it 404'd and the
client parsed the error body as float data.
Include the model in both the id and the URL, and reject a resampled grid that
contains non-finite values instead of caching it. Verified against live data:
the IFS grid now decodes to 65,160 finite values (u -20.2..27.6 m/s) and renders
a field distinct from GFS.
Live browser QA of the Wind layer surfaced several rendering defects that
unit tests with fakes did not catch:
- The trail fade used a translucent black fill (source-over), which raises
alpha instead of erasing; combined with additive strokes the overlay
saturated into a solid white disk. Fade with destination-out instead, and
draw strokes with normal blending.
- Particles were seeded uniformly over the whole globe, so at regional zoom
almost none fell on screen. Seed inside the camera's visible rectangle.
- A non-finite particle position (polar divide-by-zero) reached Cesium,
which threw and aborted the entire frame before any stroke. Guard finite
coordinates per particle and per projection.
- Exaggerate advection in proportion to camera height, so the same visual
speed is legible at street level and at hemispheric zoom.
- Brighter colour ramp (the old ramp started near-invisible dark grey) and a
defensive frame try/catch so one bad frame cannot kill the loop.
Verified: the wind field/colour/particle model renders the real GFS 06z field
correctly in an offline render, and the wind unit tests pass (8/8).
Add ECMWF IFS 10 m wind alongside NOAA GFS, selectable at runtime.
The /api/wind proxy becomes a model dispatcher: `?model=gfs|ifs` (default gfs,
unknown → 400), cached per model+cycle+grid. IFS reads the keyless ECMWF Open
Data JSON Lines `.index` (note: `<stem>.index`, not `<grib2>.index`), byte-range
fetches only the `10u`/`10v` messages, decodes them with the same ecCodes WASM
path, and resamples with the existing grid code (ECMWF grids start at lo1=180;
the resampler already wraps columns).
The client layer gains a `model` parameter (setParams/getParams) with a
generation guard so a model switch aborts and refetches exactly once. Registered
as an enabled+options layer with a `model` enum.
Validated end-to-end against the real 2026-09-14 06z IFS cycle on Node 25
(180x91 grid at 2°, 131,040-byte payload, plausible mean |u|/|v|), and GFS
still passes the same check.
Attribution: contains modified Copernicus/ECMWF IFS forecast data, licensed
under CC BY 4.0. Verified: npm test (3412 pass, 0 fail), check:boundaries.
Add a Wind layer animating the global 10 m wind field from NOAA GFS.
The /api/wind proxy selects the latest available 0.25° cycle, reads its .idx
inventory, byte-range fetches only the UGRD/VGRD 10 m GRIB2 messages (~1 MB
each instead of the whole ~500 MB file), decodes them with ecCodes (WASM),
resamples to a compact grid (1° default), and serves a manifest plus a Float32
U/V payload cached per cycle for an hour (single-flight, last-good on failure).
The client renders nullschool-style particles in a canvas overlay: bilinear
field sampling with longitude wrap, per-frame advection, Cesium camera
projection, and globe-occlusion culling via EllipsoidalOccluder.
Validated end-to-end against the real 2026-09-14 06z cycle on Node 25
(180x91 grid at 2°, 131,040-byte payload, plausible mean |u|/|v|). The WASM
decoder requires Node >=24, which the repo already requires.
Registered as `wind` (token k) with attribution and docs. Verified: npm test
(3412 pass, 0 fail), npm run check:boundaries.
Toggling datacenters, dams, or submarine cables off only set `show = false`
on their data sources. Cesium's DataSourceDisplay still walks every
visualizer of a hidden source each frame, and the ~11.5k hidden entities
(hundreds of MB) stayed resident, so the whole scene rendered ~2x slower
for the rest of the session after a single on/off round trip (owner field
test 2026-09-13).
disable() now removes the data sources from the viewer and drops the built
records; the parsed dataset (NDJSON features / TeleGeography JSON) is cached
so enable() rebuilds without a refetch or re-parse. A disable that lands
while a load is in flight releases the finished build instead of parking it
hidden. destroy() clears the cache.
Measured with a headful Chrome harness (flights+military+satellites on,
camera rotating, 250-frame windows), before -> after:
- entities resident after OFF: 11,537 -> 0
- DataSourceDisplay.update per frame after OFF: 1.09 ms -> 0.02 ms
- scene.render per frame after OFF: 13.9 ms (control 6.6 ms) -> 5.2 ms
(fresh baseline 5.0 ms in the same run)
- JS heap 30 s after OFF: 1.6 GB and climbing -> 342 MB after GC
Tests: the GeoJSON lifecycle and cables suites assert disable leaves zero
sources/contexts and re-enable rebuilds with fetch called once.
Daniel Slay